Core Concepts
Hypergraph
Understanding hypergraphs and why they matter for AI memory
Hypergraph
HyperMemory stores your agent’s knowledge in a hypergraph — a data structure that captures relationships more naturally than traditional graphs.
Graphs you already know
You’re probably familiar with regular graphs: nodes connected by edges.
Alice ────manages────► Project Alpha
│ │
│ ─────┘
└───collaborates───► BobEach edge connects exactly two nodes. This works well for simple relationships, but real-world knowledge is messier.
What makes a hypergraph different
In a hypergraph, a single edge — called a hyperedge — can connect three or more nodes simultaneously.
┌─────────────────────────────────────────────┐
│ Sprint Planning │
│ (hyperedge) │
│ │
│ Alice ─── Project Alpha ─── Sprint 12 │
│ │ │ │ │
│ └───────────┴──────────────┘ │
└─────────────────────────────────────────────┘This single hyperedge captures that Alice, Project Alpha, and Sprint 12 are all connected through the concept of “Sprint Planning” — without needing to create six separate binary edges.
Why this matters for AI memory
Real-world knowledge rarely fits into neat pairs:
- Decisions involve multiple people, options, and outcomes
- Events connect participants, locations, times, and topics
- Projects span teams, milestones, dependencies, and stakeholders
With binary edges, your agent has to reconstruct complex contexts by traversing many connections. With hyperedges, the context is already captured in one relationship.
Example: Remembering a decision
Binary graph approach:
Decision ──involves──► Sarah
Decision ──involves──► Marcus
Decision ──concerns──► API v2
Decision ──occurred_at──► Q3 Planning
Decision ──outcome──► Approved5 edges. Your agent has to traverse all of them to understand the full context.
Hypergraph approach:
[Sarah, Marcus, API v2, Q3 Planning, Approved]
↑
decision_context1 hyperedge. The full context is captured atomically.
HyperMemory’s hypergraph features
| Feature | Description |
|---|---|
| Nodes | Individual memories (facts, people, concepts, decisions) |
| Edges | Binary relationships between two nodes |
| Hyperedges | Multi-way relationships connecting 3+ nodes |
| Typed relationships | All edges and hyperedges have labels/types |
| Metadata | Attach key-value data to nodes and edges |
| Temporal awareness | Every element has timestamps; query by time range |
When to use edges vs hyperedges
| Use Case | Recommendation |
|---|---|
| Simple relationship (A manages B) | Edge |
| Category membership (A is-a type) | Edge |
| Sequential connection (A then B) | Edge |
| Complex context (meeting with 5 participants) | Hyperedge |
| Decision involving multiple factors | Hyperedge |
| Event with multiple participants/outcomes | Hyperedge |
You don’t always need to decide upfront. Start with simple edges, and create hyperedges when you recognize patterns that involve multiple connected concepts.